A multi-sensor fire detection method based on long short-term memory network and environmental information fusion

Abstract

The application discloses a kind of multi-sensor fire detection methods based on long short-term memory network and environmental information fusion.The method is when detecting fire, first using multi-sensor system continuously acquires temperature, smoke concentration, carbon monoxide concentration and carbon dioxide concentration data in environment, constructs the time series of monitoring value corresponding to environmental index;Second, based on long short-term memory network, the environmental change degree extraction module and the environmental level extraction module in the deep environmental information extraction model are constructed, with the help of long short-term memory network processing ability to long-distance dependence relationship of time series data, the change degree of environmental index and the level of environmental index are predicted;Then, using multilayer perceptron, the obtained various environmental information is fused, and comprehensive fire state information is obtained;Finally, the comprehensive fire state information value is compared with the best threshold value learned in system operation, and fire judgment is completed.

Description

Technical Field

[0001] This invention relates to the fields of fire detection and machine learning applications, and more specifically, to a multi-sensor fire detection method based on long short-term memory networks and environmental information fusion. Background Technology

[0002] Fire, as a complex combustion process, is accompanied by changes in various environmental indicators, such as temperature, smoke concentration, and the intensity of harmful gases. Traditional single-sensor fire detection methods compare the current sensor reading with a preset threshold to trigger an alarm for a single environmental indicator. However, due to the limited environmental information acquired, single-sensor detection methods are prone to false alarms and missed alarms. Subsequent fire detection methods based on multi-sensor systems can simultaneously monitor multiple environmental indicators and analyze, process, synthesize, and fuse the acquired information based on certain criteria, thereby achieving more accurate fire assessments. Multi-sensor detection methods overcome the uncertainties and limitations of single sensors, improve the overall system performance, and the redundancy of data between sensors enhances system reliability. However, most current multi-sensor fire detection methods only consider the current environmental indicator monitoring values, ignoring the deeper environmental information corresponding to these indicators, such as the change in environmental indicators quantifying the degree of change in environmental information from the previous moment, and the level of environmental indicators measuring the environmental level over time. Therefore, existing multi-sensor fire detection methods are susceptible to false alarms due to non-fire factors, such as short-term sensor malfunctions and transient electromagnetic interference. Furthermore, multi-sensor systems require processing a large amount of environmental information, resulting in complex overall structures and low detection efficiency. Therefore, how to comprehensively consider environmental indicator monitoring values, the degree of change in environmental indicators, and the level of environmental indicators, and how to efficiently achieve environmental information fusion, are important problems that multi-sensor fire detection systems currently need to solve. Summary of the Invention

[0003] The purpose of this invention is to address the shortcomings of existing technologies by providing a multi-sensor fire detection method based on long short-term memory networks and environmental information fusion. This method utilizes the excellent time series prediction capabilities of long short-term memory networks to extract deep environmental information from time series data of different environmental indicators. Combined with information fusion technology, the comprehensive fire status at the current moment is obtained for fire judgment. This method effectively avoids interference caused by non-fire factors and significantly improves the robustness and accuracy of fire detection.

[0004] To achieve the above objectives, the present invention adopts the following technical solution:

[0005] Step S1: The multi-sensor system continuously collects data on temperature, smoke concentration, carbon monoxide concentration, and carbon dioxide concentration in the environment, constructs time series of monitoring values ​​for the corresponding environmental indicators, and performs standardization processing.

[0006] Step S2: Construct a deep environmental information extraction model based on a long short-term memory network, including an environmental change degree extraction module and an environmental level extraction module;

[0007] Step S3: Use the long short-term memory network in the deep environmental information extraction model to process the time series data of the input environmental indicator monitoring values, and predict the degree of change of the environmental indicators at the current time relative to the previous time and the level of the environmental indicators at the current time.

[0008] Step S4: To improve system operating efficiency, the multilayer perceptron is pre-trained using another set of original fire data under similar conditions to determine the optimal number of hidden layer nodes for the multilayer perceptron.

[0009] Step S5: Use a multilayer sensor to fuse various environmental information at each moment in time sequence to obtain the comprehensive fire status information corresponding to that moment;

[0010] Step S6: Compare the comprehensive fire status information value with the optimal threshold learned during system operation. If it is less than the threshold, it is considered that no fire has occurred; otherwise, it is considered that a fire has occurred.

[0011] The environmental change degree extraction module in step S2 is a Long Short-Term Memory (LSTM) network. After processing through forgetting and updating gates, the cell states used for transmission are the first type of fire state information; the hidden states calculated and output by the output gate are the environmental indicator change degree information. The core idea of ​​the environmental change degree extraction module is to maintain the transmission of the first type of fire state in the neural network chain. Through the unique three-gate structure of the LTM network, the first type of fire state information is forgotten, updated, and output during the transmission process, thereby controlling the output of the environmental indicator change degree information.

[0012] The first type of fire status information is a quantitative description of the monitored fire environment based on environmental changes, and its quantitative formula is as follows:

[0013]

[0014] In the above formula (1), This represents the first type of fire state of environmental indicator n at time k. and Let these represent the forgetting vector and the update vector representing the change in environmental indicator n at time k, respectively. This represents the first type of fire status information for environmental indicator n at time k-1. Let represent the vector of candidate values ​​for the first type of fire state of environmental index n at time k; where and The expression is as follows:

[0015]

[0016] In formula (2) above, σ represents the Sigmoid activation function, W vf Indicates the forgetting weight due to environmental changes. This indicates the degree of change of environmental indicator n at time k-1. b represents the difference in monitoring data for environmental indicator n at times k and k-1. vf This indicates a forgetting bias due to changing environmental conditions.

[0017]

[0018] In the above formula (3), W vu Indicates that the weights are updated according to environmental changes, b vu The environmental changes update the bias, and the other symbols are expressed in the same way as in formula (2);

[0019]

[0020] In formula (4) above, tanh represents the tanh activation function, W v and b v These represent the weights and biases for creating the candidate value vector of the first type of fire state, respectively, and the other symbols are expressed in the same way as in formula (2).

[0021] The environmental indicator change information is used to predict the degree of environmental change at the current time compared to the previous time in a time series, and its calculation formula is as follows:

[0022]

[0023] In the above formula (5), This indicates the degree of change of environmental indicator n at time k. The output vector represents the change of environmental indicator n at time k, and tanh represents the tanh activation function. This represents the first type of fire state of environmental indicator n at time k; where, The expression is as follows:

[0024]

[0025] In formula (6) above, σ represents the Sigmoid activation function, and Wv vo The output weights represent environmental changes. This indicates the degree of change of environmental indicator n at time k-1. b represents the difference in monitoring data for environmental indicator n at times k and k-1. vo This indicates the output bias due to environmental changes.

[0026] The environmental indicator level extraction module in step S2 is another type of long short-term memory network. In this network, after processing through forgetting and updating gates, the cell states used for transmission are the second type of fire state information; the hidden states calculated and output by the output gate are the environmental indicator level information. The core idea of ​​the environmental indicator level extraction module is to maintain the transmission of the second type of fire state in the neural network chain, and to control the output and transmission of the environmental indicator level information by forgetting, updating, and outputting the second type of fire state information through a gate structure.

[0027] The second type of fire status information is a quantitative description of the monitored fire environment based on environmental levels, and its quantitative formula is as follows:

[0028]

[0029] In the above formula (7), This represents the second type of fire state of environmental indicator n at time k. and Let represent the horizontal forgetting vector and the horizontal update vector of environmental indicator n at time k, respectively. This represents the second type of fire state at time k-1, where environmental indicator n is the fire state. Let represent the vector of candidate values ​​for the second type of fire state of environmental index n at time k; where and The expression is as follows:

[0030]

[0031] In formula (8) above, σ represents the Sigmoid activation function, W lf Indicates the environmental level forgetting weight. This represents the level information of environmental indicator n at time k-1. b represents the monitoring value of environmental indicator n at time k. lf This indicates environmental level forgetting bias;

[0032]

[0033] In the above formula (9), W lu b represents the environmental level update weight. lu The symbol represents the environmental level update bias, and the other symbols are expressed in the same way as in formula (8);

[0034]

[0035] In the above formula (10), tanh represents the tanh activation function, W l and b l These represent the weights and biases for creating the candidate value vector for the second type of fire state, respectively, and the remaining symbols are expressed in the same way as in formula (8);

[0036] The environmental indicator level information is a measurement of the environmental level at the current moment over time, and its calculation formula is as follows:

[0037]

[0038] In the above formula (11), This represents the level information of environmental indicator n at time k. Let tanh represent the horizontal output vector of environmental indicator n at time k, and let tanh represent the tanh activation function. This represents the second type of fire state of environmental indicator n at time k; where, The expression is as follows:

[0039]

[0040] In the above formula (12), σ represents the Sigmoid activation function, W lo This indicates the output weights at the environmental level. This represents the level information of environmental indicator n at time k-1. b represents the monitoring value of environmental indicator n at time k. lo This indicates the environmental level output bias.

[0041] The information fusion in step S5 involves using computer and neural network technology to analyze and synthesize multi-source fire monitoring values ​​obtained in a time sequence under certain criteria. In a multi-sensor fire detection system, the environmental indicator information provided by each information source has a certain degree of uncertainty. Fusion of this uncertain information is an uncertainty reasoning process. Neural networks, with their self-learning, adaptive, and nonlinear processing capabilities, can meet the requirements of multi-sensor data fusion technology. The classification criteria are determined based on the similarity of samples received by the current system; this determination method is mainly reflected in the weight distribution of the network. The function expression for multi-layer perceptron fusion processing of multiple environmental information is as follows:

[0042]

[0043] In the above formula (13), Let W represent the output vector of the input layer of the multilayer perceptron at time k, σ represent the sigmoid activation function, and W represent the output vector of the input layer of the multilayer perceptron at time k. in and b in These represent the weights and biases of the input layer, respectively. This represents the transpose of a vector group consisting of the monitored value of environmental indicator n at time k, information on the degree of change of the environmental indicator, and information on the level of the environmental indicator.

[0044] The Sigmoid function is used to activate the output vector of the input layer. The activation expression is as follows:

[0045]

[0046] In the above formula (14), Let σ represent the output vector of the hidden layer of the multilayer perceptron at time k, and let Whi id and b hid These represent the weights and biases of the hidden layer, respectively.

[0047] Furthermore, by activating the hidden layer output vector, the comprehensive fire status is obtained, calculated using the following formula:

[0048]

[0049] In the above formula (15), F k Let W represent the overall fire state at time k, σ represent the Sigmoid activation function, and W represent the overall fire state at time k. out and b out These represent the weights and biases of the output layer, respectively. Attached Figure Description

[0050] Figure 1 This is a system architecture diagram of the present invention.

[0051] Figure 2 This is a structural diagram of the environmental indicator change degree extraction module of the present invention.

[0052] Figure 3 This is a structural diagram of the environmental indicator level extraction module of the present invention.

[0053] Figure 4 This is a structural diagram of the environmental information fusion module of the present invention. Detailed Implementation

[0054] The present invention will be further described below with reference to the accompanying drawings and embodiments, but this is not intended to limit the scope of the invention.

[0055] Example:

[0056] Figure 1This is a system architecture diagram of the present invention, which consists of an environmental indicator monitoring value processor, a deep environmental information extraction model, and an information fusion module. The multi-sensor system continuously collects environmental indicator monitoring values ​​to construct corresponding time-series data. The environmental indicator change degree extraction module and the environmental level extraction module in the deep environmental information extraction model use a long short-term memory network to process the input environmental indicator time-series data, combining the cell state and hidden state from the previous time step transmitted in the network to output change degree information and level information corresponding to different environmental indicators at the current time. Next, the information fusion module analyzes and processes various environmental information at the current time to obtain comprehensive fire status information for fire judgment. Specifically, the environmental information corresponding to each type of environmental indicator includes three parts: environmental indicator monitoring value, environmental indicator change degree information, and environmental indicator level information.

[0057] Specifically, to illustrate the implementation of the present invention, in the embodiments, the original time series data constructed from the collected environmental indicator monitoring values ​​is shown in Table 1:

[0058] Table 1. Raw time series data of monitoring values ​​for different environmental indicators (partial)

[0059]

[0060]

[0061] In this example, the monitored environmental indicators include temperature, smoke concentration, carbon monoxide concentration, and carbon dioxide concentration. A relative ignition time of 10 represents the time when a fire occurred; less than 10 represents the time before a fire occurred; and greater than or equal to 10 represents the time when a fire had occurred. The raw time-series data of the above environmental indicators are standardized using the following formula:

[0062]

[0063] In the above formula (1), This represents the standardized monitoring value of environmental indicator n at time k. This represents the original monitoring value of environmental indicator n at time k. This represents the maximum raw monitoring value of environmental indicator n over time series T.

[0064] The standardized environmental indicator monitoring values ​​are shown in Table 2:

[0065] Table 2. Time series data of different environmental indicators after standardization (partial)

[0066]

[0067] Furthermore, the standardized environmental indicator time series data are input into the deep environmental information extraction model, including the environmental indicator change degree extraction module and the environmental indicator level extraction module.

[0068] Figure 2 This invention relates to an environmental indicator change level extraction module structure. As a long short-term memory network, after processing through forget and update gates, the cell state used for transmission is the first type of fire state information; the hidden state calculated and output by the output gate is the environmental indicator change level information. The network inputs are the difference between the environmental indicator monitoring data at the current time and the previous time, as well as the environmental indicator change level information at the previous time, and then calculates the environmental indicator change level information at the current time. Both the first type of fire state information and the environmental indicator change level information obtained during the extraction process are transmitted to the next cell unit. The quantification formula for the first type of fire state is as follows:

[0069]

[0070] In the above formula (2), This represents the first type of fire state of environmental indicator n at time k. and Let these represent the forgetting vector and the update vector representing the change in environmental indicator n at time k, respectively. This represents the first type of fire state of environmental indicator n at time k-1. Let represent the vector of candidate values ​​for the first type of fire state of environmental index n at time k; where and The expression is as follows:

[0071]

[0072] In formula (3) above, σ represents the Sigmoid activation function, W vf Indicates the forgetting weight due to environmental changes. This indicates the degree of change of environmental indicator n at time k-1. b represents the difference in monitoring data for environmental indicator n at times k and k-1. vf This indicates a forgetting bias due to environmental changes;

[0073]

[0074] In the above formula (4), W vu Indicates that the weights are updated according to environmental changes, b vu The environmental change update bias is indicated by the symbol, and the rest are expressed in the same way as in formula (3);

[0075]

[0076] In formula (5) above, tanh represents the tanh activation function, W v and b v These represent the weights and biases for creating the candidate value vector of the first type of fire state, respectively, and the remaining symbols are expressed in the same way as in formula (3);

[0077] Information on the degree of change in environmental indicators is used to predict the degree of environmental change at the current time point compared to the previous time point in a time series. The calculation formula is as follows:

[0078]

[0079] In the above formula (6), This indicates the degree of change of environmental indicator n at time k. The output vector represents the change of environmental indicator n at time k, and tanh represents the tanh activation function. This represents the first type of fire state of environmental indicator n at time k; where, The expression is as follows:

[0080]

[0081] In formula (7) above, σ represents the Sigmoid activation function, W vo The output weights represent environmental changes. This indicates the degree of change of environmental indicator n at time k-1. b represents the difference in monitoring data for environmental indicator n at times k and k-1. vo This indicates the output bias due to environmental changes.

[0082] Figure 3 This invention relates to an environmental indicator level extraction module structure, which, as an alternative long short-term memory network, uses a forgetting gate and an update gate to transmit cell states representing second-type fire state information. The hidden state, calculated and output by the output gate, represents environmental indicator level information. The network inputs are the standardized environmental indicator monitoring values ​​at the current moment and the environmental indicator level information from the previous moment, thereby calculating the environmental indicator level information at the current moment. Both the second-type fire state information and the environmental indicator level information obtained during the extraction process are passed to the next cell unit. The quantification formula for the second-type fire state is as follows:

[0083]

[0084] In the above formula (8), This represents the second type of fire state of environmental indicator n at time k. and Let represent the horizontal forgetting vector and the horizontal update vector of environmental indicator n at time k, respectively. This represents the second type of fire state at time k-1, where environmental indicator n is the fire state. Let represent the candidate value vector of the second type of fire state information for environmental index n at time k; where and The expression is as follows:

[0085]

[0086] In the above formula (9), σ represents the Sigmoid activation function, W lf Indicates the environmental level forgetting weight. This represents the level information of environmental indicator n at time k-1. b represents the monitoring value of environmental indicator n at time k. lf This indicates environmental level forgetting bias;

[0087]

[0088] In the above formula (10), W lu b represents the environmental level update weight. lu This indicates the environmental level update bias; the other symbols are expressed in the same way as in formula (9).

[0089]

[0090] In formula (11) above, tanh represents the tanh activation function, W l and b l These represent the weights and biases for creating the candidate value vector for the second type of fire state, respectively, and the remaining symbols are expressed in the same way as in formula (9);

[0091] Environmental indicator level information is a measurement of the environmental level at the current moment over time, and its calculation formula is as follows:

[0092]

[0093] In the above formula (12), This represents the level information of environmental indicator n at time k. Let tanh represent the horizontal output vector of environmental indicator n at time k, and let tanh represent the tanh activation function. This represents the second type of fire state of environmental indicator n at time k; where, The expression is as follows:

[0094]

[0095] In the above formula (13), σ represents the Sigmoid activation function, W lo This indicates the output weights at the environmental level. This represents the level information of environmental indicator n at time k-1. b represents the monitoring value of environmental indicator n at time k. lo This indicates the environmental level output bias.

[0096] Figure 4 This is a structural diagram of the environmental information fusion module of the present invention, used to analyze and fuse multi-source fire monitoring values ​​obtained in a time sequence under certain criteria. The function expression is as follows:

[0097]

[0098] In the above formula (14), Let W represent the output vector of the input layer of the multilayer perceptron at time k, σ represent the sigmoid activation function, and W represent the output vector of the input layer of the multilayer perceptron at time k. in and b in These represent the weights and biases of the input layer, respectively. This represents the transpose of a vector group consisting of the monitored value of environmental indicator n at time k, information on the degree of change of the environmental indicator, and information on the level of the environmental indicator.

[0099] The Sigmoid function is used to activate the output vector of the input layer. The activation function expression is as follows:

[0100]

[0101] In the above formula (15), Let W represent the output vector of the hidden layer of the multilayer perceptron at time k, σ represent the sigmoid activation function, and W represent the output vector of the hidden layer at time k. hid and b hid These represent the weights and biases of the hidden layer, respectively.

[0102] Furthermore, by activating the hidden layer output vector, the comprehensive fire status is obtained, calculated using the following formula:

[0103]

[0104] In the above formula (16), F k Let W represent the overall fire state at time k, σ represent the Sigmoid activation function, and W represent the overall fire state at time k. out and b out These represent the weights and biases of the output layer, respectively.

[0105] The test results of the example are shown in Table 3. The test results are good and accurate.

[0106] Table 3. Actual test results of the system

[0107]

[0108] The above description is only a preferred embodiment of the present invention. All equivalent changes and modifications made within the scope of the claims of the present invention should be included in the scope of the present invention.

Claims

1. A multi-sensor fire detection method based on long short-term memory networks and environmental information fusion, characterized in that, Includes the following steps: Step S1: The multi-sensor system continuously collects data on temperature, smoke concentration, carbon monoxide concentration, and carbon dioxide concentration in the environment, constructs time series of monitoring values ​​for the corresponding environmental indicators, and performs standardization processing. Step S2: Construct a deep environmental information extraction model based on a long short-term memory network, including an environmental change degree extraction module and an environmental level extraction module; Step S3: Use the long short-term memory network in the deep environmental information extraction model to process the time series data of the input environmental indicator monitoring values, and predict the degree of change of the environmental indicators at the current time relative to the previous time and the level of the environmental indicators at the current time. Step S4: Use another set of original fire data under similar conditions to pre-train the multilayer perceptron and determine the optimal number of hidden layer nodes for the multilayer perceptron; Step S5: Use a multilayer sensor to fuse various environmental information at each moment in time sequence to obtain the comprehensive fire status information corresponding to that moment; Step S6: Compare the comprehensive fire status information value with the optimal threshold learned during system operation. If it is less than the threshold, it is considered that no fire has occurred; otherwise, it is considered that a fire has occurred.

2. The multi-sensor fire detection method based on long short-term memory network and environmental information fusion according to claim 1, characterized in that: The environmental change degree extraction module in step S2 is a long short-term memory network. After processing by the forget gate and the update gate, the cell state used to transmit is the first type of fire state information; the hidden state calculated and output by the output gate is the environmental indicator change degree information.

3. The multi-sensor fire detection method based on long short-term memory network and environmental information fusion according to claim 2, characterized in that: The first type of fire status information is a quantitative description of the monitored fire environment based on environmental changes, and its quantitative formula is as follows: In the above formula (1), This represents the first type of fire state of environmental indicator n at time k. and Let these represent the forgetting vector and the update vector representing the change in environmental indicator n at time k, respectively. This represents the first type of fire state of environmental indicator n at time k-1. Let represent the vector of candidate values ​​for the first type of fire state of environmental index n at time k; where and The expression is as follows: In formula (2) above, σ represents the Sigmoid activation function, W vf Indicates the forgetting weight due to environmental changes. This indicates the degree of change of environmental indicator n at time k-1. b represents the difference in monitoring data for environmental indicator n at times k and k-1. vf This indicates a forgetting bias due to environmental changes; In the above formula (3), W vu Indicates that the weights are updated according to environmental changes, b vu The environmental changes update the bias, and the other symbols are expressed in the same way as in formula (2); In formula (4) above, tanh represents the tanh activation function, W v and b v represents the weight and bias of creating the candidate value vector for the first type of fire state, respectively, and the other symbols are expressed in the same way as in formula (2).

4. The multi-sensor fire detection method based on long short-term memory network and environmental information fusion according to claim 2, characterized in that: The environmental indicator change information is used to predict the degree of environmental change at the current time compared to the previous time in a time series, and its calculation formula is as follows: In the above formula (5), This indicates the degree of change of environmental indicator n at time k. The output vector represents the change of environmental indicator n at time k, and tanh represents the tanh activation function. This represents the first type of fire state of environmental indicator n at time k; where, The expression is as follows: In formula (6) above, σ represents the Sigmoid activation function, W vo The output weights represent environmental changes. This indicates the degree of change of environmental indicator n at time k-1. b represents the difference in monitoring data for environmental indicator n at times k and k-1. vo This indicates the output bias due to environmental changes.

5. The multi-sensor fire detection method based on long short-term memory network and environmental information fusion according to claim 1, characterized in that: The environmental level extraction module in step S2 is a long short-term memory network. After processing by the forget gate and the update gate, the cell state used to transmit is the second type of fire state information; the hidden state calculated and output by the output gate is the environmental indicator level information.

6. The multi-sensor fire detection method based on long short-term memory network and environmental information fusion according to claim 5, characterized in that: The second type of fire status information is a quantitative description of the monitored fire environment based on environmental levels, and its quantitative formula is as follows: In the above formula (7), This represents the second type of fire state of environmental indicator n at time k. and Let represent the horizontal forgetting vector and the horizontal update vector of environmental indicator n at time k, respectively. This represents the second type of fire state at time k-1, where environmental indicator n is the fire state. Let represent the vector of candidate values ​​for the second type of fire state of environmental index n at time k; where and The expression is as follows: In formula (8) above, σ represents the Sigmoid activation function, W lf Indicates the environmental level forgetting weight. This represents the level information of environmental indicator n at time k-1. b represents the monitoring value of environmental indicator n at time k. lf This indicates environmental level forgetting bias; In the above formula (9), W lu b represents the environmental level update weight. lu The symbol represents the environmental level update bias, and the other symbols are expressed in the same way as in formula (8); In the above formula (10), tanh represents the tanh activation function, W l and b l represents the weight and bias of creating the candidate value vector for the second type of fire state, respectively, and the other symbols are expressed in the same way as in formula (8).

7. The multi-sensor fire detection method based on long short-term memory network and environmental information fusion according to claim 5, characterized in that: The environmental indicator level information is a measurement of the environmental level at the current moment over time, and its calculation formula is as follows: In the above formula (11), This represents the level information of environmental indicator n at time k. Let tanh represent the horizontal output vector of environmental indicator n at time k, and let tanh represent the tanh activation function. This represents the second type of fire state of environmental indicator n at time k; where, The expression is as follows: In the above formula (12), σ represents the Sigmoid activation function, W lo This indicates the output weights at the environmental level. This represents the level information of environmental indicator n at time k-1. b represents the monitoring value of environmental indicator n at time k. lo This indicates the environmental level output bias.

8. The multi-sensor fire detection method based on long short-term memory network and environmental information fusion according to claim 1, characterized in that: In step S5, a multilayer perceptron is used to fuse multiple environmental information at each moment in the time series. The environmental information corresponding to each type of environmental indicator includes the monitored value at that moment, the degree of change of the environmental indicator at that moment relative to the previous moment, and the environmental indicator level at that moment. The fusion function is as follows: In the above formula (13), Let W represent the output vector of the input layer of the multilayer perceptron at time k, σ represent the sigmoid activation function, and W represent the output vector of the input layer of the multilayer perceptron at time k. in and b in These represent the weights and biases of the input layer, respectively. This represents the transpose of a vector group consisting of the monitored value of environmental indicator n at time k, information on the degree of change of the environmental indicator, and information on the level of the environmental indicator; the output vector of the input layer is activated using the Sigmoid function, and the activation expression is as follows: In the above formula (14), Let W represent the output vector of the hidden layer of the multilayer perceptron at time k, σ represent the sigmoid activation function, and W represent the output vector of the hidden layer at time k. hid and b hid These represent the weights and biases of the hidden layer, respectively. Activate the hidden layer output vector to obtain the comprehensive fire status, calculated using the following formula: In the above formula (15), F k Let W represent the overall fire state at time k, σ represent the Sigmoid activation function, and W represent the overall fire state at time k. out and b out These represent the weights and biases of the output layer, respectively.

Images